mlsanityData to 7 Oct 2026

Membership Inference Attack Against Music Diffusion Models via Generative Manifold Perturbation

arXiv 2602.01645 ↗First posted 2 Feb 2026cs.SDcited by 1

On the map: Diffusion models → Robustness, privacy and adversarial aspects

Abstract (arXiv)

Membership inference attacks (MIAs) test whether a specific audio clip was used to train a model, making them a key tool for auditing generative music models for copyright compliance. However, loss-based signals (e.g., reconstruction error) are weakly aligned with human perception in practice, yielding poor separability at the low false-positive rates (FPRs) required for forensics. We propose the Latent Stability Adversarial Probe (LSA-Probe), a white-box method that measures a geometric property of the reverse diffusion: the minimal time-normalized perturbation budget needed to cross a fixed perceptual degradation threshold at an intermediate diffusion state. We show that training members, residing in more stable regions, exhibit a significantly higher degradation cost.

Key facts. Cited by 1 paper in the map (data through 7 Oct 2026). Placed in Robustness, privacy and adversarial aspects, one of 14 directions in Diffusion models. Builds on 19 of its 22 references that are in the map. First posted on arXiv 2 Feb 2026.

Builds on

19 of 22 references.

Closest ancestors · 5

Same direction, newest first

  1. 2024Towards Black-Box Membership Inference Attack for Diffusion Models25 May
  2. 2023White-box Membership Inference Attacks against Diffusion Models11 Aug
  3. 2023An Efficient Membership Inference Attack for the Diffusion Model by Proximal Initialization26 May
  4. 2023Membership Inference Attacks against Diffusion Models7 Feb
  5. 2023Extracting Training Data from Diffusion Models30 Jan

Other references · 14

  1. 2020Denoising Diffusion Probabilistic Modelscited by 20,851
  2. 2016Membership Inference Attacks against Machine Learning Modelscited by 2,221
  3. 2020DiffWave: A Versatile Diffusion Model for Audio Synthesiscited by 1,339
  4. 2016On Large-Batch Training for Deep Learning: Generalization Gap and Sharp Minimacited by 1,337
  5. 2022Memorization Without Overfitting: Analyzing the Training Dynamics of Large Language Modelscited by 218
  6. 2018Enabling Factorized Piano Music Modeling and Generation with the MAESTRO Datasetcited by 215
  7. 2016FMA: A Dataset For Music Analysiscited by 205
  8. 2020Membership Leakage in Label-Only Exposurescited by 156
  9. 2019Reconstruction and Membership Inference Attacks against Generative Modelscited by 122
  10. 2023MusicLDM: Enhancing Novelty in Text-to-Music Generation Using Beat-Synchronous Mixup Strategiescited by 94
  11. 2022Membership Inference Attacks by Exploiting Loss Trajectorycited by 93
  12. 2024MeLFusion: Synthesizing Music from Image and Language Cues using Diffusion Modelscited by 19
2 more
  1. 2025Membership Inference Attacks on Sequence Modelscited by 6
  2. 2025Generative AI Training and Copyright Lawcited by 3

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 1.

1 paper.

  1. 2026Auditing Training Data in Generative Music Models via Black-Box Membership Inferencecited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2023Are Diffusion Models Vulnerable to Membership Inference Attacks?
  2. 2026Noise as a Probe: Membership Inference Attacks on Diffusion Models Leveraging Initial Noise
  3. 2026When Noise Lowers The Loss: Rethinking Likelihood-Based Evaluation in Music Large Language Models
  4. 2025Enhancing Membership Inference Attacks on Diffusion Models from a Frequency-Domain Perspective
  5. 2023Towards More Realistic Membership Inference Attacks on Large Diffusion Models
  6. 2026Black-box Membership Inference Attacks on the Pre-training Data of Image-generation Models
  7. 2026Where Flow Matching Leaks: Characterising Membership Signals Along the Interpolation Path
  8. 2024Model Will Tell: Training Membership Inference for Diffusion Models